The selection of roost cavities by the threatened New Zealand long-tailed bat Chalinolobus tuberculatus was examined in unmodified temperate southern beech Nothofagus rainforest in Fiordland, New Zealand. We radio-tracked 73 bats during spring–autumn of 1993–1997 to 155 day roosts, all of which were in trees. One hundred and forty-nine roosts were in cavities: 84 in live trunks, 33 in dead trunks, 32 in large branches (seven in dead branches, 25 in live); and the remaining six (occupied by solitary bats) were beneath loose bark. We compared characteristics of the 84 roost cavities in live trunks with 57 random available cavities also in live trunks. Cavities used by C. tuberculatus were not a random subset of available cavities. In comparison to available cavities, bat roosts were all located in knot hole cavities, were high from the ground, and had little surrounding vegetation. All roosts were dry inside. Bats used a higher proportion of cavities which had medium sized entrances and internal cavities, and thicker cavity walls. A logistic regression model incorporating nine cavity variables classified roost and available cavities correctly 97% and 91% of the time, respectively. Distance to the nearest vegetation, cavity condition (wet or dry inside), height from the ground, and to a lesser extent internal cavity height, explained significant proportions of the variation between roost and available cavities. Bats changed roost sites virtually every day and it appeared that suitable cavities were abundant in the unlogged forests of the study area. It is unlikely that modified forests and forests managed for timber production contain such a large pool of potential roosting cavities.